More systematic handling of `library`, `attach`, etc. shimming

[brodieG]

Seems like those functions change more than anticipated. Breaks in 3.2, and now in 3.3. At a minimum seems like we need reference versions of them for older R versions.

Is there a strategy to work around this issue? How terrible is it to check search path before and after each statement? Obviously we'll miss out on any statements that attach/detach or load/unload within a single statement, but we can only observe the final result after the statement is executed anyway, so is there really a huge advantage to tracking every search path / namespace change? Options?

Actually, the problem is that our special environment that bypasses .GlobalEnv needs to be updated any time library, etc. are called otherwise top level test expressions won't see the newly attached packages.

We could provide our own library, etc functions to try to address this. Should be okay because only top level tests are affected. Any package functions that risk calling library/etc. directly should be relying on NAMESPACE. That said, we add a high risk of difficult to debug problems.

[brodieG]

Fixed by switching to a new tracing mechanism that traces at end of those functions.